This code refers to https://github.com/hzy46/fast-neural-style-tensorflow.
You can see the middle result in the output/
And you can see the finally result in the generated/
Last version exists the Checkerboard Artifacts because I use the tf.nn.conv2d_transpose
And you can see the https://distill.pub/2016/deconv-checkerboard/ to understand about Checkerboard Artifacts
It works still not good,and I think because it just train the three images for 1000 ecophs
Train data 5000 steps about 20000 images the result become better
